On Fri, 28 Oct 2005, Chuck Swiger wrote:

> Gavan Fantom wrote:
>> Todd Vierling wrote:
> [ ... ]
>>> If a system has many GiB of memory, it should be up to the admin to lower
>>> the execmin in order to gain yet more fs or anon page cache.  It should
>>> *not* be up to the users of systems with modest memory to raise them.
>> 
>> This says to me that the variables themselves could do with a rethink, not 
>> just the default values. Perhaps there's something other than a percentage 
>> that makes for a better constant default?
>> 
>> How does the ideal size of execmin (and, perhaps the other mins) vary with 
>> total system memory size?
>
> It depends on load and the variety of programs being run.  A machine like a 
> mailserver or a DNS server might only run a small number of programs, whereas 
> an end-user workstation running KDE and lots of eyecandy might want an order 
> of magnitude more room for text pages.
>
> [ Delurks long enough to mutter something about looking at the global page 
> fault frequency rate of the various page types, and trying to minimize that. 
> Easier said than coded. :-) ]

 	Everytime this comes around someone usually makes a comment about
 	Solaris having some nice page-stealing algorithm whereby the
 	chance of one type of memry user stealing a page from another
 	is inversely related to the percentage used by the stealing type.

 	Which in English means the larger the number of file pages the
 	less chance a new filepage request will take a page from anyone
 	else.

 	For example if you have 80% filepages and 10% execpages then
 	a request for a filepage probably has 1% chance of taking an
 	exec page, whereas a request for an exec page has 99% of taking
 	a filepage.

 	-
-- 
 		David/absolute       -- www.NetBSD.org: No hype required --